Goals: The goal is to develop a computational system to predict the targets toward which small molecules are biased. The tool will optimize biased libraries, prioritize them for testing against particular targets, identify targets in phenotypic screens, and allow users to easily move from compounds to targets. Significance: The tool would allow investigators to easily identify small molecules to modulate a target. It also would allow them to build libraries that are biased toward such targets. Users of the tool include investigators looking to quickly access biological molecules for their targets, companies interested in building libraries biased toward such targets, and vendors looking to expand the usefulness of libraries that they sell. A motivation is a tool that can rapidly bring chemical matter to biologists. Theory/Background: We and others have shown that it is possible to predict previously unknown "off- targets" for drugs. Our Similarity Ensemble Approach (SEA) uses chemical structure to predict targets for any molecule by associating that molecule with similar patterns found in ligands annotated to protein targets. Whereas many commercially available molecules have no target linkage, on average for any given vendor we predict that about T of them do. Aim 1. To develop a service to assign targets to small molecules in large libraries. We will build a system that will allow users to query a library of small molecules, asking one of two questions. a. For a given target, what small molecules in this library would be expected to modulate it? b. For a given library, on what targets are its molecules most likely to work? Milestone: The essential features of this system exist, and proof of principle has been demonstrated in predicting new targets for over 25 drugs using SEA. Here we broaden the method to enable use by non- experts on large compound libraries. There are two pragmatic milestones. i. Experimentally testing predicted compound-target associations to demonstrate feasibility. ii. Development of a web-interface that can be integrated with vendor catalogs. Aim 2. To develop a service that can optimize a library for target coverage. a. Given a particular library size, can we optimize it to cover the maximum number of pharmacologically relevant targets? b. Correspondingly, can we optimize a library for the maximum target bias? Milestone: This method is essentially in hand, but has not been tested. We will work with our commercial partners to optimize their library for target coverage, demonstrating this by predicting and experimentally testing molecules for target activity (six months). Aim 3. To predict the targets for compounds active in phenotypic and animal assays. A recent development in pharmaceutical research and chemical biology has been the rediscovery of phenotypic, even whole organism screens for compound activity, frequently using modified cells or organisms that carry a known disease-associated genotype. A core challenge is identifying the molecular targets responsible for the observed phenotype. Indeed, our partners in pharma indicate that they can have hundreds of related compounds with activity in cells, tissues, organs or animal models, but do not know the actual molecular targets, limiting mechanistic understanding and optimization. We will use SEA to predict targets for these active molecules. Milestone: In collaboration with a pharmaceutical partner, we will predict targets for a compound series with animal model activity, but for which targets remain unknown. Initially we anticipate testing ten such compounds in receptor-binding assays. Whereas these goals are ambitious, extensive preliminary results suggest that they are feasible. PUBLIC HEALTH RELEVANCE: Investigators often test large numbers of compounds to discover a chemical starting point for a new drug. In this proposal we focus on providing a tool to predict which commercially-available compounds are most likely to be active at therapeutic targets. This saves time and money by reducing the number of compounds investigators need to test during early-stage drug discovery.